# Feature Loop Scheme

> Full feature development workflow (called by feature-new/feature-continue)

- Skill: `majiayu000/feature-loop-scheme` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add majiayu000/feature-loop-scheme`
- Raw SKILL.md: https://api.skillmd.com/api/skills/majiayu000/feature-loop-scheme/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: majiayu000 (https://skillmd.com/u/majiayu000)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/majiayu000/feature-loop-scheme

---


**Prerequisites:** Must be in a feature clone directory.

## Process

### Step 1: Gather Context
Use the explorer subagent to gather context about the codebase relevant to the feature:
- Explore existing code patterns and architecture
- Identify related files and components
- Understand dependencies and integration points
- Set thoroughness level to "medium" for balance between speed and depth

This context will inform the planning phase.

### Step 2: Plan or Analyze
Determine feature name from branch: `FEATURE_NAME=$(git rev-parse --abbrev-ref HEAD)`
- **If plan_$FEATURE_NAME.md doesn't exist** → Create plan using planner subagent with opus model, ask questions if needed
- **If plan_$FEATURE_NAME.md exists** → Analyze current progress compared to origin/main, examine plan_$FEATURE_NAME.md and documentation, identify next steps

### Step 2.5: Commit Plan (if just created)
If plan was just created in Step 2, commit and push it immediately so it survives session crashes:
```bash
git add plan_$FEATURE_NAME.md && git commit -m 'Add feature plan' && git push
```

### Step 3: Create Task List
Create a structured task list based on the plan or next steps:
- Break down into actionable tasks
- Use the TaskCreate tool to create the task list
- Each task should be specific and measurable
- Mark the first task as "in_progress" to begin work

### Step 4: Implement
- Code, commit, push with coder-agent
- If problems occur, fix, commit, push with debugger-agent

### Step 5: Quality
Run quality skill to fix code style, types, and remove AI slop.

### Step 6: Review
Use Task tool with subagent_type="reviewer-agent" for final code review and validation.

### Step 7: Summary
Report findings and confirm ready for PR (or list remaining issues).